Types of Crosslinguistic Variation in Case Assignment
Mark Baker
Rutgers University
In this work, I do not address directly issues about what range of hypotheses concerning
crosslinguistic variation are or aren’t Minimalist, or questions about micro- vs. macroparameters, or clarifying the role of the lexicon versus syntactic principles in
crosslinguistic variation, or so on. It is not because I’m above talking about such things or
don’t have opinions about them. That is simply not what I want to focus on here. Rather
I want to offer something more along the lines of addressing what Cedric Boeckx calls
Greenburg’s problem. I agree with him that Greenburg’s problem is quite distinct from
Plato’s problem, but I think that it’s also interesting, and we should be trying to come up
with solutions for it—although not perhaps in exactly the way Greenburg might have
foreseen.
In particular, I’ll focus here on a specific issue in syntactic variation, namely the
theory of morphological case, especially overt structural case. I imagine that if there is
such a thing as abstract Case (i.e. NP licensing) that that is systematically related, but that
topic is not my focus here. In presenting this research, I also want to illustrate a style of
relating to the material that I want to recommend, what I’ve been calling Formal
Generative Typology (Baker, 2010). This involves committing to two kinds of ideas
which I think are both attractive. The first is the generativist vision of accepting a degree
of abstractness in our descriptions of languages, and looking at how insight into empirical
details can emerge from proper formalization of the patterns. The second is the idea of
Greenbergian typology, that it can be valuable to compare radically different crosslinguistic systems. I’ll try to combine these two threads within case theory, to see what
kinds of things might be out there, and to pose in a preliminary way the question of what
might be the most interesting things to say about the variations that we find. Once that
has been done, we can speculate about whether the results are truly Minimalist or not.
This work breaks down into two large parts. In the first (sections 1 and 2), I
review two recent studies of mine that have bearing on how you might think about crosslinguistic variation in morphological case. In the second part (sections 3-6), I give some
first thoughts about what might emerge of we use those studies as wedges into the more
general issue of cross-linguistic variation in this area. The second part will be somewhat
preliminary, but hopefully we will feel like we have learned something by approaching
the topic in this way by the end.
1. Implications of a typology of agreement for case assignment
The first seed of a new approach to case comes from my formal generative typology of
agreement systems in Baker (2008). It emerged as a result of this study that languages
differ significantly in the relationship between agreement and case. In Chomsky’s (2000,
2001) vision, agreement and case are two sides of the same coin. And indeed, there is a
tight relationship between subject agreement and nominative case in Indo-European (IE)
languages, where the two track each other quite closely. It is reasonable, then, to
1
conclude that these are two morphological realizations of a single abstract agreement
relationship (Agree). But one result of my study was that this is true for a large class of
languages, but it is parameterized in some sense; it is a point of cross-linguistic variation.
Indeed, I argued that it was a macroparameter: it is variation in the syntactic operation of
Agree, and not in the feature content of individual functional heads involved in
agreement.
The kind of data that first led me to think along these lines involved differences
between Niger-Congo languages and IE languages. Languages from both families have
subject agreement. So, you can say something like “the women chopped wood with an
ax” and the finite verb “chop” agrees with the subject “women” in both a Bantu language
like Kinande (see (1)) and an IE language like Catalan.1
(1)
Abakali
mo-ba-seny-ire
olukwi
women.2
AFF-2S/T-chop-EXT wood.11
‘The woman chopped wood (with an axe).’
(lw’-omo-mbasa).
LK11-LOC.18-axe.9
(Kinande)
However, if you look more closely you find that there are two senses of the term
“subject” in play here. In one set of languages, the crucial notion of subject is Spec,TP.
That seems to be what the finite verb agrees with in most Bantu languages. In contrast, in
IE languages what we call subject agreement is really agreement with the thematic
subject, maybe, or (better) the NP that gets the nominative case. One is a case-oriented
principle, the other is a position-oriented principle. You can see the difference when you
consider inversion constructions, like those in 2 and 3.
(2)
kw-a-hir-aw-a
ehilanga.
Oko-mesa
LOC.17-table 17S-T-put-PASS-FV peanuts.19
‘On the table were put peanuts.’
(Locative Inversion)
(Kinande)
(3)
si-lu-li-seny-a
bakali
(omo-mbasa). (Kinande)
a. Olukwi
wood.11 NEG-11S-PRES-chop-FV women.2 LOC.18-axe.9
‘WOMEN do not chop wood (with an axe).’
(Object fronting)
b. (I believe that) no good things does/*do he withhold from those who walk…
2 is a locative inversion sentence. Here you didn’t put the highest argument, or the thing
with nominative case, in the Spec,TP position, so the two senses of subject come apart. In
1
Glosses used in the abbreviations include: ABS, absolutive; ACC, accusative: AFF, affirmative;
AGNOML, agentive nominalizer; AOR, aorist; ASP, aspect; AUX, auxiliary; CN, connector; COM,
comitative; DAT, dative; ERG, ergative; EXT, aspect extension; F, feminine; FUT, future; FV, final vowel;
GEN, genitive; IMPF, imperfective; INEL, inellesive; INSTR, instrumental; LK, linker; LOC, locative; M,
masculine; NEG, negative; NOM, nominative; OBL, oblique; PASS, passive; PAST, past tense; PERF,
perfective; PL, plural; PN, proper noun; PRES, present; PTPL, participle; SG, singular; T, tense.
Agreement affixes are glossed with a triple symbol, starting with a number indicating the person (1, 2, or
3), a lower case letter indicating number (s or p), and an upper case letter expressing the grammatical
function/case (S, O, A, E, P) of the agreed with nominal. Other numbers in the glosses of Kinande
examples indicate noun classes.
2
English and in most other IE languages, the verb agrees with “peanuts” in the inversion
sentence. “Peanuts” is in nominative case, it’s arguably the highest argument, and that’s
what you agree with. In contrast, in the Bantu languages the verb agrees with the fronted
element “on the table” (see also Bresnan and Kanerva 1989). You also see something
similar in languages which allow you to move the object to Spec,TP position. This is less
common in both language families, but it happens in some. In Kinande, you can put the
thematic object “wood” in the Spec,TP position to get 3a. Now the finite verb in Kinande
agrees with “wood”, not with “women”. There are arguably some IE languages where a
similar kind of object fronting happens, such as Yiddish and Icelandic, as well as
negative inversion in English. The exact details of these fronting processes may not be
identical in the various languages. But the crucial thing is that in any such inversion
structures in IE the verb still agrees with the nominative subject, as in 3b: “No good
things” is plural, but “does” is singular, matching the postverbal NP “he” instead. So
there’s a different agreement pattern in IE as opposed to Bantu. This led me to propose 4
as a statement of the relevant parameter:
(4)
The Case Dependence of Agreement Parameter (CDAP): (Baker 2008)
F agrees with DP/NP only if F values the Case feature of DP/NP (or vice versa).
(No: most Niger Congo languages; Yes: most IE languages)
Saying that a functional head F agrees with a noun phrase only if F values the case
feature of that noun phrase is a statement that comes out of the Chomsky Agree theory. I
concur that it’s a property of grammar, but it’s a parameterized property, not a universal
one; it holds of some linguistic systems and not others.
I went on to claim that the CDAP in 4 is a macroparameter. First of all, it seems to
hold of all functional heads in the relevant languages, not just one particular functional
head (see Baker 2008:ch. 5).2 Second, it has other kinds of grammatical consequences
that you might not have noticed at first. Another consequence concerns how agreement
happens in constructions consisting of a main verb together with one or more auxiliary
verbs. Here too there is difference between Niger-Congo languages and IE languages,
shown in 5 versus 6.
(5)
a. Abakali ba-bya ba-ka-gul-a
amatunda.
women.2 2S-were 2S-PTPL-buy-FV fruits.6
‘The women were buying fruits.’
b. Tú-lwé
(Kinande)
tú-ká-ly-a.
2
Luigi Rizzi points out that person agreement in Italian is related to (nominative) case assignment, but
number-gender agreement on adjectives or participles is not. He suggests then that the CDAP might have
different values for different heads in the same language. In Baker 2008, I argued for a different
interpretation: I claimed that agreement is always contingent on there being a case relationship in IE
languages, but the necessary case relationship can be one of case concord as well as one of case
assignment. Indeed, in richly-inflected Icelandic it is clear that adjectives which agree with an NP in case
also agree with it in number and gender, whereas adjectives that cannot agree with an NP in case (because
that NP has quirky case) do not agree with it in number and gender either. This is my reason for including
the phrase “or vice versa” in 4. A fully-developed theory of case concord remain to be worked out,
however.
3
1pS-leave 1pS-PTPL-eat-FV
‘We were eating.’
(6)
a. *John is comes.
(English)
b. Nosotros estamos leyendo el libro.
we
are(1pP) reading the book
(*leyendemos)
(reading-1pS)
(Spanish)
Bantu languages often allow multiple full agreement in these constructions. For example,
in 5a you have the same number-gender agreement ba- on both “were” and “buy”. This
extends even to person agreement, as shown in 5b, where the first person plural prefix tuappears on both verbs. You don’t find this kind of full multiple agreement in IE, as
shown in 6. Typically in IE you have agreement on the auxiliary, but not the main verb.
Depending on the language, you might also have a reduced adjectival-like agreement on
the main verb under some circumstances, but not full person agreement. I claimed that
this was another effect of the CDAP. Suppose that the CDAP is set “yes”. Then the fact
that you can’t have multiple agreement with the same NP follows from the fact that you
can’t have multiple case marking of that NP. Imagine that the first head agrees with the
NP. That presupposes that it assigned case to that NP. An NP can only be case-marked
once (I assume). Then if another head tries to agree with the NP, it can’t because the NP
has already been case-marked. Therefore multiple agreement with the same argument
can’t arise. However, that follows crucially because we assumed that there’s a
relationship between case-marking and agreement. If there is no such relationship, as in
Bantu languages, then the conclusion doesn’t hold and nothing rules out agreeing with
the same NP more than once. That’s the source of the difference between 5 and 6, I
claim.
Any claims about case in Bantu languages are a bit abstract because case is not
morphologically realized in these languages. I assumed above that that the fronted object
in 3a is accusative, but that could be debated. Fortunately, you see the same parametric
difference in languages that have overt case marking. In this domain too you find two
classes of languages: languages where morphological case marking is closely correlated
with the agreement and languages where it isn’t. Hindi is a language in which there is a
close connection, given familiar examples like those in 7 (Mohanan, 1995:83).
(7)
kitaabẽ
becegaa
a. Anil
Anil.M(NOM) book-F.PL sell-FUT.M.SG
‘Anil will sell (the) books.’
b. Anil-ne kitaabẽ
becĩĩ
Anil-ERG book-F.PL sell-PERF-F.PL
‘Anil sold (the) books.’
c. Anil-ne kitaabō-ko
becaa
Anil-ERG book-F.PL-ACC sell-PERF-M.SG
‘Anil sold the books.’
4
Hindi is a split ergative language. In 7a the subject is in nominative case, but in 7b, in a
different different-aspect, the subject is in ergative case. This difference impacts the
agreement. When the subject is in nominative case, as in 7a, the verb agrees with the
subject. When the subject is in ergative case in 7b, the verb doesn’t agree with it; rather it
agrees with a nominative case object. In 7c where neither the subject nor the object is
nominative, then the verb has to be a default third person form. Case and agreement are
closely interrelated here. But there are also languages like Burushaski, a language spoken
in the Himalayas, where you find just the opposite (data from Lormer 1935). Burushaski
has case marking similar to Hindi, but the T of the finite verb mindlessly agrees with the
subject, regardless of its case. 8a has a nominative case subject, and the verbs agree with
it. 8b has an ergative case subject and the verbs still agree with it. There’s no relevant
morphological difference on the verbs.
(8)
a. Jε
u:ņε xidmʌt
εč-a
b-a.
I.NOM your service
do-1sS(IMPF) be-1sS
‘(For these many years) I have been at your service.’
b. Ja
be.ʌdʌpi.εn εt-a
b-a.
I.ERG discourtesy do-1sS(PERF) be-1sS
‘I have committed a discourtesy.’
Something very similar is true of object agreement as well. According to standard
Chomskyan theory, object agreement should be related to accusative case assignment. 9a
shows object agreement on the verb with the object “you” in absolutive case. 9b shows
that you have the same morpheme “gu” agreeing with an object bearing a dative case.
(9)
gu-yεtsʌ-m.
a. (U:ņ)
you.ABS 2sO-see-1sS
‘I saw you.’
b. U:ņ-ər
hik trʌņ gu-čičʌ-m.
you-DAT one half 2sS-give-1sS
‘I shall give a half to you.’
Hindi then is a CDAP=yes language, Burushaski a CDAP=no language. Notice also that
the Burushaski examples in 8 have multiple person agreement: the main verb “do” is first
person singular, and the auxiliary verb “be” is also first person singular, both agreeing
with the subject. In this respect also Burushaski is like Kinande, whereas Hindi is like
other IE languages.
When I did this research, I was studying agreement, not case. Nevertheless, we
can follow up by asking what this parameterization in the relationship between agreement
and case tells us about case. What are the sources of case marking across languages?
Having the CDAP set “yes” implies that case is coming from the agreement-bearing
functional categories. But about languages like Burushaski or Kinande where case is not
related to agreement? Does case come from some different kind of source in such
5
languages? Does my agreement parameter imply that case theory is also more
parameterized than we have realized?
2. Modalities of case assignment in Sakha (Turkic)
With these questions in mind, I next review the highlights of a language-particular study
that I did with Nadya Vinokurova on case assignment in Sakha, a Turkic language
spoken in Siberia (Baker and Vinokurova, 2010). The hope here is that if you take
knowledge about how case assignment works in one language and combine it with a
picture of how agreement works across languages, you might be able to project this into
some new ideas about how case works across languages. In particular, I’m going to
compare nominative case in Sakha to accusative case in the same language. I’ll show that
nominative case is related to agreement on T in the familiar way. But accusative case is
not related to agreement with a functional head. Instead I claim that the rules for
accusative case assignment are quite different in nature from the rules of nominative
assignment.
Consider first nominative case and its relationship to agreement. I claim that
Sakha is a CDAP=yes language, like IE, so this part looks familiar. More specifically,
the two heads that agree overtly in Sakha, finite T and possessive D, both respect the
CDAP. I only discuss finite T here, in the interest of space. Examples 10 and 11 show
that finite verbs in this language cannot agree with an NP that has case other than
nominative. 10 is a minimal pair from the passive construction (Vinokurova, 2005:336):
(10)
a. Sonun-nar aaq-ylyn-ny-lar.
new-PL read-PASS-PAST-3pS
‘The news was read.’
b. Sonun-nar-y
aaq-ylyn-na.
new-PL-ACC read-PASS-PAST.3sS
‘The news was read.’
(*aaq-ylyn-ny-lar)
read-PASS-PAST-3pS
In the Sakha passive, the theme argument can be nominative, as in 10a. But it can also be
accusative, as it is in 10b. (I return to this below.) Note that the agreement on the verb is
also different in the two examples: in 10a, it agrees with “news”; in 10b it does not, but
has to be a default, third person singular form. Example 11 adds data from a dative
subject construction. This shows that you can’t have plural agreement with a dative
subject—just as you cannot in Icelandic, for example.
(11)
Oqo-lor-go üüt naada-(*lar).
child-PL-DAT milk need-(*3pS)
‘The children need milk.’
So we see that in Sakha the finite verb agrees with nominative NPs, but not NPs with any
other kind of case, like Hindi but unlike Burushaski. Sakha is a language with casesensitive agreement.
6
Example 12 shows another consequence of this parameter setting, involving
agreement in auxiliary verb constructions. If we are right to treat Sakha as a CDAP=yes
language, then we expect that there should not be multiple agreement with the subject in
such constructions. 12 shows that that’s correct. More specifically, 12a shows that you
can put person agreement on the auxiliary verb. 12b shows that you can put person
agreement on the main verb instead—an option that IE doesn’t allow. But 12c shows that,
unlike Kinande or Burushaski, you can’t have agreement on both verbs. So the cluster of
properties associated with the “yes” setting of the CDAP is holding together nicely.
(12)
a. En süüj-büt
e-bik-kin
you win-PTPL AUX-PTPL-2sS
‘(The result is that) you won.’
b. En süüj-bük-kün e-bit
you win-PTPL-2sS AUX-PTPL
c. *En süüj-bük-kün e-bik-kin
you win-PTPL-2sS AUX-PTPL-2sS
Based on data like this, Baker and Vinokurova (2010) end up with 14 as the rule
of nominative case assignment in Sakha. This is just all the fine print from Chomsky’s
(2000, 2001) theory of Agree brought together in a single statement.
(13)
If a functional head F ∈ {T, D} has unvalued phi-features and an NP X has an
unvalued case feature [and certain locality conditions hold], then agreement
happens between F and X, resulting in the phi-features of X being assigned to F
and the case associated with F (nominative or genitive) being assigned to X.
But now let’s shift attention to accusative case. Where does that come from? The
standard Minimalist view has been to generalize the story of where nominative case
comes from to accusative. We say that there’s another functional head, small v or
something similar, that agrees with an NP in lower down in the structure, the object. That
relationship is spelled out as accusative on the object, and as agreement on the functional
head. However, this relationship need not be spelled out; maybe the postsyntactic
vocabulary insertion happens not to put in forms for this agreement, even though it
happened in the syntax in the normal way. But I claim that that’s not the right rule for the
assignment of accusative case in Sakha. It’s not a coincidence that we do not see object
agreement overtly here; that’s telling us something grammatically significant in this
particular language. I argue instead for the formulation in 14, based on Marantz 1991.
Marantz introduced the notion of what’s called dependent case assignment. His view was
that case doesn’t necessarily come from a functional head; rather you look at domains,
and if you have two NPs in a particular local domain you assign accusative case to the
lower one. What’s distinctive about this is it’s a relationship between two noun phrases,
7
without the functional heads coming into it per se (except perhaps to help define the
domains). 14 is a concrete version of this leading idea.3
(14)
If there are two distinct argumental NPs in the same phase such that NP1
c-commands NP2, then value the case feature of NP2 as accusative unless
NP1 has already been marked for case.
In defense of 14, I’ll briefly go through three bits of evidence in Sakha, all of which
support arguments of approximately the same form: whether you have accusative case
depends on whether a subject is around and doesn’t depend on the nature of the
functional heads that are nearby.
First, 15 shows again that in a Sakha passive the theme argument can be
nominative or accusative. What is the grammatical difference between the two? There is
no morphological difference on the verb that we can see; both verbs bear the same
passive suffix. But there is an interpretative difference, pointed out by Vinokurova
(2005). There’s a sense in which the agent is syntactically present in 15b that it isn’t in
15a. This becomes evident if you include adverbs that presuppose an agent—adverbs like
“intentionally”, or instrumental expressions like “with a hammer”. If such elements are
included, then the thematic object “the cup” can only be accusative:
(15)
a. Caakky (*sorujan)
(*ötüje-nen)
aldjat-ylyn-na.
cup
intentionally hammer-INSTR break-PASS-PAST.3sS
‘The cup was broken (intentionally) (with a hammer).’
b. Caakky-ny sorujan
ötüje-nen
aldjat-ylyn-na.
cup-ACC intentionally hammer-INSTR break-PASS-PAST.3sS
‘The cup was broken intentionally with a hammer.’
I assume this means that 15b has some kind of empty category subject in the syntax, and
15a doesn’t, and the empty category is involved in licensing these adverbs in some way.
Now if 15b has a null subject in the syntax, then there’s a second NP in the local domain
that triggers accusative case on the theme by the dependent case rule in 14. In contrast,
15a has no independent evidence for a covert agent NP in the syntactic representation.
Therefore, there’s nothing to trigger dependent case assignment on the theme argument
and that argument shows up as nominative in 15a. So there’s no detectable difference in
the functional categories associated with the verb in these two sentences. There is a
detectable difference in whether an agent is present or not, however. And whether the
agent is present or not determines whether accusative is assigned or not. That’s a success
3
Andrea Moro asks whether 14 would apply to predicate nominal constructions, assigning accusative to the
predicate NP. That would be wrong for most languages, including Sakha, where the predicate nominal is
nominative/unmarked. In fact, though, 14 does not apply if the predicate nominal does not count as
“argumental”. I included this qualification to prevent null expletive NPs from triggering accusative in
examples like 20, and to prevent bare NP adverbs like ‘here’ and ‘now’ from being accusative in Sakha, but
it could cover this situation too. We will, however, have to be alert to this issue when it comes to
crosslinguistic variation, since how “borderline” NPs count for 14 may be a point of variation. (Expletives
may trigger accusative on other nominals in Amharic, for example (Amberber 2005), and NP adverbs are
marked accusative in ways that suggest 14 in Quechua (Lefebvre and Muysken 1988).)
8
for the Marantzian view of dependent case assignment, as contrasted with the
conventional view that accusative case comes from a particular functional category.4
Another structure where you can see this is agentive nominalizations like 16.
(16)
Terilte-ni
salaj-aaccy
kel-le.
company-ACC manage-AGNOML come-PAST.3sS
‘The manager of the company came.’
Here we have a deverbal nominal “manager”, and its theme argument “company” has
accusative case on it. This is clearly different from languages like English, which do not
allow an object in accusative case in comparable structures. The question then is: Where
does this accusative case come from? Why is it possible in Sakha, but not in familiar
Indo-European languages? My claim is that it’s the rule of accusative case assignment
that is different, not the structure of the agentive nominalizations. So, in these kinds of
nominalizations in Sakha, there is no evidence for any verbal functional heads. You can’t
have any voice marker on the nominalized verb, you can’t have any aspect markers, you
can’t have any negation, and you can’t have any adverbs, just as you can’t in English.
This is shown in part in 17 (see Baker and Vinokurova 2009 for more data).
(17)
a. (*Ücügejdik) terilte-ni
(*ücügejdik) salaj-aaccy
kel-le.
(*well)
company-ACC (*well)
manage-NOM come-3sS
‘The one who manages the company well came.’
(no adverb)
b. *tal-yll-aaccy
choose-PASS-AGNOML
‘the one who is chosen, the be-chosen-er’
(no voice morphology)
So there’s no sign of any clausal structure in these constructions, except the verb and its
complement. That’s a real problem if you think accusative case comes by agreement with
a clausal functional head; there would be no source for accusative here. That is arguably
a good result for English, but a bad one for Sakha. There is however a syntactic
expression of the agent in 16, namely the agentive nominalizer itself. So, if you take the
dependent case view where it’s having two nominal elements in the same domain that’s
crucial for the assignment of case, then you can generalize that rule to this structure with
relatively few difficulties. (The main issue is that one has to define “argumental NP” in
14 in such a way that the nominalizing suffix –aaccy counts as one.)
Perhaps the strongest argument for 14 comes from the examples in 18-20. These
illustrate an interesting kind of raising to object that you find in Sakha. Like in Turkish
(Moore, 1998), but not identical to Turkish, there is a kind of exceptional case marking
even with finite clauses in this language. 18a shows a simple embedded clause, and the
4
Richard Kayne points out that the difference between 15a and 15b in Sakha might be syntactically
analogous to the difference between a true passive and a (reflexive) impersonal construction in Italian,
constructions that have distinct morphological marking. If so, the similarity in verbal morphology might be
semi-accidental. I accept that this is a possibility, but do not think it would detract much from the point at
hand: even if the voice marker is technically different in 15a and 15b, the examples still point toward there
being a direct relationship between having a syntactic subject (possibly covert) and having an accusative
object.
9
embedded subject “you” has nominative case, as you would expect. But, if you raise
“you” to the edge of this clause, putting it before all the material in the lower clause, then
it can be marked accusative, as in 18b.
(18)
a. Min
[sarsyn ehigi-(*ni) kel-iex-xit
dien] ihit-ti-m.
I(NOM) tomorrow you-(*ACC) come-FUT-2pS that
hear-PAST-1sS
‘I heard that tomorrow you will come.’
b. Min [ehigi-ni [bügün kyaj-yax-xyt dien]] ihit-ti-m.
I
you-ACC today win-FUT-2pS that hear-PAST-1sS
‘I heard that you will win today.’
This fact by itself doesn’t distinguish between the two kinds of case assignment. You can
get this result by dependent case assignment, or by case coming from a functional head in
the matrix. Indeed, it would be pretty much the same story either way: when you raise the
NP up, it comes to be in the domain of the main clause, so it can get accusative case in
the matrix clause, either by agreement with the matrix v, or by 14. What’s more revealing
is the other conditions on this case assignment. Suppose the matrix verb is intransitive
rather than transitive, what happens? In 19, the matrix verbs “become sad” and “return”
are unaccusative verbs that do not have a transitive v. These verbs cannot license
accusative case in a simple one-clause sentence. They’re the kind of verbs that we would
say don’t have the functional head that assigns accusative case. But, if you raise the
embedded subject from the complement of this sort of verb, it still gets accusative case:
(19)
a. Keskil Aisen-y
[t kel-be-t
dien] xomoj-do.
Keskil Aisen-ACC
come-NEG-AOR that become.sad-PAST.3sS
Keskil became sad that Aisen is not coming. (p. 366)
b. Masha Misha-ny
[t yaldj-ya
dien] tönün-ne.
Masha Misha-ACC
fall.sick-FUT that
return-PAST.3sS
‘Masha returned (for fear) that Misha would fall sick.’
“Become sad” doesn’t have the right kind of functional head to assign accusative, but
there is a second NP in the matrix clause domain. When we raise the subject from the
embedded clause to the matrix clause, now there are two NPs in the matrix domain.
Accusative is assigned to the lower one, according to the rule in 14.
Further proof comes from comparing an example like 20, where there isn’t a
second argumental NP in the matrix clause, because the matrix clause contains an
impersonal verb “it became certain”. 20 shows that if you raise the embedded subject into
this kind of matrix clause, it does not get accusative case.
(20)
Bügün
munnjax-xa [Masha-(*ny) [ehiil
Moskva-qa
bar-ya
today
meeting-DAT Masha-(*ACC) [next.year Moscow-DAT go-FUT.3sS
dien]] cuolkajdan-na.
that]
become.certain-PAST.3sS
‘It became clear today at the meeting that Masha will go to Moscow next year.’
10
The contrast between 19 and 20 shows that whether or not there’s a subject in the matrix
clause is crucial to accusative case assignment. At the same time, the lack of contrast
between 18 and 19 shows that the nature of the matrix verb and the functional categories
associated with it is not crucial to whether one gets accusative case in Sakha.
Putting together evidence like this, I conclude that Sakha is a language where
accusative case comes from a rule of dependent case assignment, not one where
accusative case is a manifestation of an agreement relationship with some dedicated
functional category. When you control the structure to eliminate the functional category,
accusative case is still there.
This is a very language specific study. But it is typologically significant because it
gives us good reason to believe that there’s a richer range of different kinds of case
assignment available in languages of the world than most of us have thought.
Suppose then that we try to project this forward into a more general theory of how
case can vary across languages. Section 1 (from Baker 2008) showed that case comes
from functional heads in some languages (CDAP=yes languages) but not others
(CDAP=no languages). Section 2 (from Baker and Vinokurova 2010) showed that even
in one language (Sakha), there’s two kinds of case assignment: case assignment by
agreement with a functional head (nominative), and case assignment by dependent case
marking—a calculation involving two NPs in the same domain. Maybe then the
language-particular study helps us to answer the question of where case comes from in
CDAP=no languages like Burushaski, or in languages with little or no sign of agreement
on functional heads, like Chinese and Japanese. Maybe it comes from Marantzian
dependent case assignment rules. If so, case in agreement-rich languages would come
from a partially different source than case in agreement-poor languages. I want to begin
considering whether that’s true or not.
More specifically, I have begun to do sketches of about twenty five non-IE
languages with overt case marking, to see how these matters play out in that sample. I’m
going for twenty five languages, not two, in the hope of achieving a fair degree of
generality. But I’m going for twenty five languages, not three hundred, so as to be able to
apply a touch of generative sophistication in the analysis of each of them. With this sort
of methodology in mind, the rest of this work presents some preliminary feasibility
studies along these lines.
3. On ergative case marking and its distribution
I start with the observation that’s always been a bit tricky to know where ergative case
systems fit in with a Chomskian approach to case assignment. So in a morphologically
ergative language you have the subject in the SpecTP position (maybe) but that doesn’t
determine its case. Even if the subject doesn’t raise to SpecTP, it is probably the closest
NP to T but that doesn’t necessarily determine its case either. Thus in Burushaski the
subjects is sometimes absolutive (21a,b), sometimes ergative (21c), even though
structural tests suggest that it is always in the same position. Sometimes the absolutive
NP agrees with T (21a,b), but sometimes it shows object agreement (21c). There’s no
tight correlation between case and either position or agreement in this language. And
11
that’s a problem for the standard Chomskian approach. There have been plenty of clever
suggestions about what to do about this, but some kind of patch seems needed.
(21)
i-ír-imi
a. Acaanák hilés
suddenly boy.ABS 3m-die-PAST.MsS
‘Suddenly the boy died.’
(Willson, 1996:19)
b. Dasín
há-e
le hurúT-umo
girl.ABS house-OBL in sit-PAST.FsS
‘The girl sat in the house.’
(Willson 1996:3)
dasin
mu-yeéts-imi.
c. Hilés-e
boy-ERG girl.ABS FsO-see-PAST.MsS
‘The boy saw the girl.’
(Willson 1996:17)
In contrast, it’s trivially easy to fit ergative case systems into the dependent case
approach. Indeed, part of Marantz’s original motivation for this was that you can account
for accusative and ergative in the same way. 22a repeats the accusative case rule from 14.
Now you get an ergative pattern out of the same rule if you simply say that when you
have two NPs in the same phase, you mark the higher one with overt case, rather than the
lower one, as in 22b. For Marantz, that’s all there is to morphological ergativity.
(22)
a. If there are two distinct argumental NPs in the same phase such that NP1
c-commands NP2, then value the case feature of NP2 as accusative.
b. If there are two distinct argumental NPs in the same phase such that NP1
c-commands NP2, then value the case feature of NP1 as ergative.
Another virtue of the Marantzian approach (not mentioned by Marantz himself) is
that it generalizes easily to so-called tripartite systems like Nez Perce, illustrated in 23.
23a has an intransitive verb, and “man” has no explicit case marking. 23b is a transitive,
and the subject has ergative case and the object has accusative case. So in this language
you have a three-way contrast: an NP can be plain (nominative/absolutive) or ergative or
accusative.
(23)
a. Hi-páay-na
háama.
3S-arrive-ASP man
‘The man arrived.’
(Rude, 1986:126)
b. Háama-nm hi-néec-‘wi-ye
man-ERG 3S-pO-shoot-ASP
‘The man shot the elk(pl).’
wewúkiye-ne.
elk-ACC
(Rude 1986:127)
So we get a simple sort of typology from the Marantz picture, in which you either mark
the higher of two NPs in the same domain ergative (Burushaski), or you mark the lower
one accusative (Sakha), or you mark both (Nez Perce), or you mark neither (Mohawk,
12
Swahili, etc.). One might say, then, that languages are parameterized to make use of 22a,
or 22b, or both, or neither.
Now, what might this imply if we say that some languages do case assignment by
the Chomskian way, and others do it by the Marantzian way? What would that suggest
about the distribution of ergativity? We know that ergativity is not evenly distributed
across the languages of the world: for example, ergativity is rare in Europe (apart from
Basque) but common in New Guinea and Australia. So this suggests the following
conjecture: In CDAP=yes languages like most of IE, case is related to agreement. But
you can’t get ergative patterns out of normal theories of agreement. Therefore, you
should not have ergative case marking in languages with that kind of system. In contrast,
in languages where you either have no agreement or where agreement is independent of
case marking, like Burushaski, then case does not come from agreement. Instead, it
comes from Marantzian dependent case rules. And in those kinds of languages you can
have ergative and accusative equally well. So the conjecture would be that CDAP=yes
languages will always be nominative-accusative (if they have overt case marking at all),
whereas CDAP=no languages and languages in which agreement is not active in the
syntax will be nominative-accusative half the time and ergative-absolutive half the time.
The table in 24 is my first attempt to test this conjecture. Here I cross my CDAP,
taken from Baker 2008, with a simple typology of whether the language is ergative, or
accusative or has no case marking at all, taken directly from The World Atlas of
Language Structures (Comrie 2005) (counting tripartite languages as ergative).
(24)
: (n=102)
Agreement is CaseSensitive (CDAP=yes)
(41)
Ergative2? (expected: 7.4)
Absolutive
(Hindi,
(18)
Greenlandics)
Nominative- 12 (expected: 11.9)
Accusative
Hebrew, Greek,
(29)
Sakha, Kannada…
No case
26 (expected: 22.6)
marking (55) English, Mapudungun,
Yimas, Hausa…
Agreement is not
Case-sensitive
(CDAP=no) (32)
12 (expected: 5.8)
Georgian, Warlpiri,
Basque, Burushaki…
7 (expected: 9.3)
Persian, Guarani, I.
Quechua, Maricopa,
14 (expected: 17.6)
Kinande, Slave,
Arapesh, Ojibwa…
Languages with no
Agreement (29)
4 (expected 5.2)
Lezgian, Ngiyambaa
10 (expected 8.4)
Malagasy, Japanese,
Marthathunira, Hiaki
15 (expected 16)
Yoruba, Mandarin,
Mixtec, …
There are two suggestive features of this table to point out. The first is that you have very
few ergative-absolutive languages with case sensitive agreement in the top left corner:
only Hindi and Greenlandic. It seems to be true, then, that languages where case is
sensitive to agreement are almost all nominative-accusative, a 6:1 ratio, as predicted. In
contrast, in the other two columns, you have quite a few ergative-absolutive languages.
Indeed, if you sum up the other two columns, you have sixteen ergative-absolutive
languages and 17 nominative-accusative languages—a 50-50 split, as predicted. This
roughly supports my conjecture. It is admittedly quite crude so far. One would like to
dive into all of these languages with more care, and especially the potentially anomalous
13
languages Hindi and Greenlandic (both of which have nontrivial generative literatures
that would be significant).5 But even these very preliminary results look quite promising.
4. Specificity of the object and dependent case marking
This proposal about ergativity makes a novel typological prediction—one that at first
glance seems troubling, but that turns out to be true. The characteristic feature of
dependent case assignment is that it assigns accusative or ergative case if and only if
there are two NPs in the same local domain. It does not apply if, for example, you have
one noun phrase way up in the matrix clause, and a second one far below it, in a triply
embedded clause. In the current theoretical climate, it makes sense to say that that the
crucial local environment is the phase in something like Chomsky’s sense. So we expect
the rule of dependent case assignment to be something of the form “If there are two noun
phrases in the same phase, then mark one of them accusative or ergative.”
For nominative-accusative languages, this looks like it has straightforwardly
positive results, accounting for a particular kind of “differential object marking” (DOM).
Indeed, Sakha itself is a DOM language: it’s a language in which the object may or may
not be accusative, depending on its position and its specificity. Thus, the direct object
“book” in 25 is marked accusative only if it is specific or definite.
(25)
a. Erel kinige-ni
atyylas-ta.
Erel book-ACC buy-PAST.3sS
‘Erel bought the/a certain book.’
b. Erel kinige atyylas-ta.
Erel book-Ø buy-PAST.3sS
‘Erel bought a book/books.’
There’s also an interaction between case and word order in Sakha, as shown in 26. The
accusative object most naturally appears in front of adverbs and other VP-internal
elements, as in 26a, whereas the bare object has to be after adverbs and other VP-internal
elements, as in 26b.
(26)
a. Masha salamaat-*(y) türgennik sie-te.
Masha porridge-ACC quickly eat-PAST.3sS
‘Masha ate the porridge quickly.’ (ACC is required)
b. Masha türgennik salamaat-(#y) sie-te.
Masha quickly porridge-ACC eat-PAST.3sS
‘Masha ate porridge quickly.’ (ACC is marked, only with focus on ‘porridge’)
5
For example, Hindi is known to have inherent/quirky ergative case (present on subjects
of some monadic verbs and missing on subjects of some dyadic verbs, based on lexical
factors cf. Woolford 2006). Greenlandic is tricky because both subject and object agree
with a single head, T (after Bok-Bennema 1991).
14
What we seem to have here is a classic case of object-shift. The object is first generated
inside VP. It can, however, move out of VP. This affects its word order with respect to
adverbs, naturally. It also affects its interpretation, assuming that Diesing’s (1992)
Mapping Hypothesis, or something like it, is correct in saying that the scope of a certain
type of interpretation is the VP. Now within the dependent case theory, it makes sense
that object shift will affect the case marking of the object as well. The VP is a distinct
phase from the clause as a whole.6 When the object NP moves out of the VP, it will be in
the same phase as the subject and therefore is marked accusative. When the object stays
in VP, it gets a different kind of interpretation and it is not in the same domain as the
subject. Then 22a does not apply, and the NP remains unmarked for case, showing up in
bare-nominative form. This is diagramed in 27.
(27)
a.
b.
TP
TP
Masha
vP
NP
ti
v´
Adv
T
v´
ti
v
NP
vP
NP
Past
NPn
Past
v´
porridge VP
v
ACC
Adv
VP
VP
quickly
phase
Masha
T
VP
T´
NPi
T´
NPi
V
quickly
porridge eat
(unmarked)
phase
NP
V
tn
eat
Now this leads to an unfamiliar prediction. If the Marantzian rules of dependent
case assignment hold in languages in which the CDAP is not set “yes”, and if those rules
can be sensitive to phase boundaries, then it follows that ergative case marking on the
subject should also depend on object shift in some languages. At the empirical level,
whether the subject is marked ergative or not should be related to where the object is in
the word order of the clause, and whether it is interpreted as specific or not. And that is a
somewhat surprising prediction. It is perhaps not surprising that features of the object,
such as whether it’s specific or not, would influence the case marking of the object itself;
that’s a very local sort of effect. But if what is really at work is object shift plus
dependent case marking, then whether the object shifts or not should also influence
whether the subject is ergative or not—a nonlocal, intrinsically relational effect. This
seems a bit counterintuitive and is not known from influential typological descriptions of
case systems (e.g. Dixon 1994, Blake 1994, Hopper and Thompson 1980).
6
See Baker and Vinokurova 2010 for discussion of vP versus VP as the smaller phase in Sakha.
15
But in fact it turns out not to be so hard to find languages with the predicted
properties. One is Ika, a Chibcan language of Columbia. Ika is an SOV language. In 28a,
the direct object “jaguar” is indefinite and inside VP, adjacent to the verb. In this
example, there’s no ergative case marking on the subject. But in 28b the object is “his
pig”, a definite object, and now the subject is marked ergative. The interpretation of the
object determines the marking of the subject, as predicted. 28c shows an object “puma”
that’s explicitly moved out of the verb phrase into the higher domain, to give OSV word
order; here too the subject is marked ergative. So, the case marking of the subject
depends on the position and interpretation of the object in Ika, exactly as predicted.
(28)
a. Gsʌriwieri tigri aʔwasa-na.
Gabriel
jaguar chase-DIST
‘Gabriel went after a jaguar.’
(Frank, 1990:115)
b. Tigri-seʔ
tšinu k -ga-na. (Frank 1990:9)
Jaguar-ERG pig
PERI-eat-DIST
‘A jaguar ate his pig.’
c. Guiadžina zʌ-gʌmmɨ perɨ-seʔ an-aʔkuss-i
guak-akí nuʔ-na
puma
GEN-child dog-ERG REF-bite-while kill-PRF AUX-DIST
The dog had killed the puma’s cub, biting it. (Frank 1990:116)
29 shows a similar paradigm from Eastern Ostyak, a Finno-Ugric language spoken in
Siberia. 29a has an indefinite object next to the verb, and the subject is not marked
ergative; 29b has a definite object shifted over a PP, and the subject is marked ergative
(Gulya, 1966).
(29)
a. Mä
t’əkäjəγlämnä
ula
mənγäləm.
We.dual(nom) younger.sister-COM berry pick-PAST-1pS
‘I went to pick berries with my younger sister.’
b. Mə-ŋən ləγə əllə
juγ kanŋa aməγaloγ.
We-ERG them large tree beside put-PAST-3pO/1pS
‘We put them (pots of berries) beside a big tree.’
Kanuri (Nilo-Saharan, spoken in Africa) is a third language that meets this description,
with ergative suffix –ye appearing on the subject only in OSV order, or sometimes if the
object is definite (Hutchison, 1981).
Nez Perce, the language with a tripartite case system, is also relevant to this
prediction. In this language, the case of both the subject and the object depend on the
definiteness of the object, because both 22a and 22b are in force. In 30a, the object is
indefinite, and neither the subject nor the object is marked for case. In 30b, the object is
definite, and both are marked for case: the subject ergative and the object accusative. One
finds either both ergative and accusative in a Nez Perce sentence, or one finds neither;
16
the two are conditioned by the same factor—the position/interpretation of the object—
just as this system expects.
(30)
hi-‘wí-ye
wewúkiye.
a. Háama
man.NOM 3S-shoot-ASP elk.NOM
‘The man shot an elk.’
(Rude, 1988:552)
b. Háama-nm hi-néec-‘wi-ye
man-ERG 3S-pO-shoot-ASP
‘The man shot the elk(pl).’
(Rude 1986:127)
wewúkiye-ne.
elk-ACC
Moralizing on this just briefly, I think it is an instructive example of the weakness
of a purely bottom-up, data-driven methodology, such as characterizes most functionalisttypological research on topics like this. That research has discovered the dependence of
object marking on object interpretation, because that is a relatively direct, obvious
dependence. But it has not discovered the dependence of subject marking on object
interpretation, even though it is out there, because that is a less obvious dependency.
People apparently did not think to look for it—and to a significant degree you only find
what you know to look for. That is why it is important to also function in a top-down,
deductive manner: theoretical reflection can give you valuable new ideas of what we
should be looking for.7
5. Tense-Aspect splits and dependent case marking
This approach also makes a converse prediction, which I’ll discuss in a brief and tentative
manner. We know from the typological literature that some of the so-called ergative
languages are really split ergative languages: whether the subject is marked ergative or
not depends on the tense-aspect of the clause. Hindi is a familiar example of this (see 7);
it is also true in Burushaski (see 8, also Willson 1996:17): the subject of a past/perfective
sentence is marked ergative, but the subject of a future/imperfective sentence is not.
Within a theory of ergative case marking built around 22b, it is tempting to analyze this
phenomenon also in terms of phase boundaries. We might say that the future tense
morpheme in Burushaski marks its complement (vP?) as being an extra phase boundary,
in addition to the normal phases CP and VP. As a result, the subject in SpecTP is in a
different phase from the object, even if the object shifts out of VP and into vP to get a
7
It is intriguing to wonder if there might be an even deeper relationship between object shift and ergative
case marking. Richard Kayne points out that people have observed that there are no strictly SVO languages
with ergative case marking (e.g. Bittner and Hale 1996)—an observation supported by the World Atlas of
Language Structures, which lists 17 ergative languages with SOV order but 0 with SVO order (Dryer 2005
plus Comrie 2005). It is tempting to incorporate this by combining my ideas about case assignment with
Kayne’s 1994 claim, rooted in antisymmetry, that SOV is always a derived order, created by leftward
movement of the object or something that contains it. So unlike SVO languages, SOV languages always
have a kind of object shift, which could feed ergative case marking. But there are issues to face along this
line too. First, one would have to distinguish two kinds of object movement, one that feeds ergative
marking and one that does not, to account for the languages mentioned in this section. More seriously, one
would also expect overt accusative case only in SOV languages given the symmetry built into 22. It is true
that overt accusative is more common in SOV languages than in SVO languages (28 languages versus 9),
but there clearly are SVO languages with overt accusative. This is an interesting area for further research.
17
definite interpretation. Given this, 22b would not apply. In contrast, the complement of
past T/perfective aspect is not a phase, so ergative case assignment would apply more
widely in this tense.
Whatever the exact details might be, if it is right to account for tense-aspect
conditioned split ergativity in something like these terms, then I predict that accusative
case marking should also be sensitive to the tense-aspect of the clause in some languages.
Again, this follows directly from the symmetry of the rules for ergative case marking and
accusative case marking. If that conception is right, then what one is sensitive to, the
other should also be sensitive to. But again, the sensitivity of accusative case assignment
to tense-aspect is not familiar from typological research.
In fact, my sample of 25 languages does not contain many examples of even
ergative case marking being conditioned by tense-aspect. I begin to wonder if this kind
of split ergativity is really an areal phenomenon, characteristic only of languages in South
and Southwestern Asia. Other than Burushaski, the one language in my sample with split
ergativity is Coast Tsimshian, an Amerindian language of British Columbia. The case
system of this language is summarized in 31. Note that the case markers associated with
common nouns show an ergative-absolutive pattern, with the special marker da used for
transitive subjects, whereas those associated with proper nouns display a tripartite
system, with distinct markers for intransitive subject, transitive subject, and object,
similar to Nez Perce.
(31)
Coast Tsimshian Case:
Common:
Proper:
absolutive
(a)
(a)s
ergative
da
dit
accusative
(a)
(a)t
32 gives sentences with common nouns, ergative case appearing on the subject of the
transitive verb in 32b. (Note that the case particle associated with the following NP
actually encliticizes to the preceding word at PF; compare Anderson 2005 on Kwakwala.)
(32)
a. Yagwa baa-[a]
wan.
PRES run-ABS deer
‘The deer is running.’
(Dunn 1995:60)
b. Yagwa-t niis-da ts’uu’ts-a laalt
PRES-3sE see-ERG bird-ACC worm
‘The bird sees the worm.’
33 is an example with a different tense, past instead of present, and in 33b the subject is
not marked as ergative, but rather absolutive, just like the object.
(33)
a. Nah siipg-a
hana’a
PAST be.sick-ABS woman
‘The woman was sick.’
b. Nah t’uus-a
PAST push-ABS
(Dunn 1995:60)
‘yuuta
hana’k
man-(ABS) woman
18
‘The man pushed the woman.’
So this qualifies as an instance of tense-aspect based split ergativity. We might then say
that the complement of nah (but not yagwa) is an extra phase, as diagrammed in 34.
(Alternatively, we might say that vP is a phase in both tenses, but nah has an EPP feature
that causes the subject to move out of the vP phase prior to case assignment, whereas
yagwa does not.)
(34)
TP
T(+AGR)
AspP
NP
Asp´
man Asp
extra phase iff
Asp=nah
vP
nah <NP>
yagwa
v´
v
VP
phase
V
push
NP
woman
.
If the object is indefinite, it doesn’t undergo object shift out of VP, but “incorporates”
into the verb. Then the subject is not marked ergative even with present tense yagwa:
(35)
nagwaat.
a. Yagwa łee-m-lak-s
PRES haul-CN-firewood-ABS.PN his.father
‘His father is hauling firewood.’
(Dunn 1995:61)
b. Yagwa sa-a’asg-as
noo-yu.
PRES make-seaweed-ABS.PN mother.my
‘My mother is picking seaweed.’
(Mulder, 1994)
On the other hand, if the object is a pronoun, then it raises all the way out of vP as well as
VP to cliticize to (the verb in) T. Then the subject is marked ergative even with past
tense nah:
(36)
a. Na-t
‘niidz-n-t
Past-3sE see-2Obj-ERG.PN
Dzon.
John
19
‘John saw you.’
(Dunn 1995:63)
b. Na-t
lu‘niis-d-it
nagwaadu.
Past-3sE stare.at-3Obj-ERG.PN my.father
‘My father was staring at them.’ (Mulder, 1994:87)
These examples confirm that it is inadequate to simply say that yagwa assigns ergative
case to the subject, but nah assigns absolutive across the board. In both tenses, the
transitive subject can be ergative or absolutive, depending on exactly where the object is.
There is a difference between the two tenses when it comes to case marking, but where
the NPs are in the structure of the clause is an equally important factor, as expected in the
dependent case theory.
The question now is whether accusative case assignment is ever sensitive to tenseaspect in the same way that ergative assignment is. We can answer this within Coast
Tsimshian by considering the proper nouns, where there is an accusative case distinct
from absolutive. And the answer turns out to be yes, as shown in 37. 37a is a present
tense sentence, with ergative marking for the subject and accusative marking for the
object. 37b is a corresponding past tense sentence. It has absolutive marking on the
subject, as expected. But it also has absolutive marking, not accusative marking, on the
object:
(37)
a. Yagwa-t t’uus-dit
Pres-3sE push-ERG.PN
‘John is pushing Mary.’
Dzon-it
Meli.
John-ACC.PN Mary.
(Dunn 1995:67)
b. Nah t’uus-as
Dzon-s
Meli.
past push-ABS.PN John-ABS.PN Mary
‘John pushed Mary.’
So the tense-aspect split in Coast Tsimshian affects ergative case marking and accusative
case marking equally, just as the definiteness of the object affects ergative and
accusative equally in Nez Perce. This supports the logic of rules like 22. I don’t yet
know of a language that is purely nominative-accusative in which accusative case varies
with tense-aspect in this way. But that might just be an accidental gap, since tense-aspect
conditioning doesn’t seem to be all that widespread anyway.
6. Variable case-marking languages versus consistent case marking languages
While the symmetries presented in the last two sections support my neo-Marantzian
approach to morphological case in languages where the CDAP is not set “yes”, they also
raise one additional question about crosslinguistic variation in case assignment. Some
languages are DOM languages, in which object shift affects the case marking of the
subject and/or the object. But other languages are not: they have uniform case marking
regardless of properties of the object. What kind of cross-linguistic variation could this
be?
20
Perhaps the most obvious thing to say would be that in some languages object
shift is only triggered by definiteness, whereas in others all objects undergo object shift.
This would be a plausible sort of minimalist analysis, in that you can build it into the
features of the v node in a familiar way. In some languages, v has an unspecified EPP
feature that can attract any kind of DP. In other languages v bears a specific feature that
only attracts +definite phrases. That would be a fairly straightforward sort of theory.
The problem is that word order doesn’t seem to bear it out. If this approach were
correct, one might expect Subject-Object-PP/Adverb-Verb order to be normal in
languages with consistent case marking. The object being regularly separated from the
verb would be evidence that objects of all sorts shift out of VP. In contrast, SubjectAdverb/PP-Object-Verb order would be normal for indefinite objects in DOM languages,
because the indefinite object does not shift out of VP in this kind of language. But this
seems to be false. Burushaski and Lezgian are languages that always have ergative case
on the transitive subject, regardless of the definiteness of the object, but still S-X-O-V
order is perfectly possible, perhaps even preferred in these languages (38a,b). Imbabura
Quechua is a language in which objects in matrix clauses are always marked accusative,
but it too favors S-X-O-V order when the object is indefinite, just as Sakha does.
(38)
a. Hilés-e
dasín-mo-r
toofá-muts píish o-t-imi.
Boy-ERG girl-OBL.F-to gift-PL
present 3pO-do-PAST.3sS
‘The boy presented gifts to the girl.’ (Burushaski, Willson 1996)
b. Gada.di wiči-n
žibin.da-j
c’akul aqud-na.
Boy.ERG self-GEN pocket-INEL feather take.out-AOR
‘The boy took a feather out of his pocket.’
(Lezgian, Haspelmath 1993)
c. Juzi Marya-man muti-ta
kara-rka.
José María-to
mote-ACC give-PAST.3sS
‘José gave/served mote María.’
(Quechua, Cole 1985:70-71)
It is conceivable that a more careful study of word order, which controlled closely for
discourse functions and information status, might reveal some sort of difference between
languages with DOM and languages without. But it does not seem likely, given the data
at hand.8
The most promising proposal from an empirical viewpoint that I am aware of
sounds distinctly unminimalist. It would be to say that in some languages, you calculate
case for the VP, you throw the VP material away, and then you calculate the case for the
rest of the clause. That is the standard view of how phases work, and it gives you a DOM
language. To get a language without DOM, you calculate case on the VP domain
8
Another alternative might be to parameterize whether VP counts as a phase or not. In a DOM language
like Sakha, VP would count as a phase, as above, but in a non-DOM language like Quechua, VP would not
count as a phase, so the subject and object would always automatically be in the same domain for 22. This
might work for some languages (like Amharic; Baker 2010), but not in general. Baker and Vinokurova
2010 claim that in some languages dative case is also a dependent case, assigned to the higher of two DPs
in a VP-phase. The proposal under consideration would then predict that no language could have dative as
a dependent case without also differential object marking. That seems seems unlikely, since Burushaski,
Lezgian, and Quechua all have rather normal dative cases (although I don’t know their properties for sure).
21
(perhaps assigning dative; see note 8), then you extend to the larger domain, but you
don’t throw the VP-internal material away. You keep that material, and it is included in
the calculation of how to assign case on the CP phase. Then when the subject comes into
view, you always have the object as well, regardless of where it is. As a result, in this sort
of language you will get ergative or accusative case marking across the board, not
depending on object shift or definiteness. Now this would admittedly be a surprising kind
of parameter. Some have been so rash as to say that what counts as a phase might vary
across languages (e.g. Fox and Pesetsky 2004), but no one has said that how phases
function in the derivation can vary across languages. But I for one do not think that we
should rule out such possiblities a priori, as being fundamentally not minimalist, given
whatever notion of minimalism we have at the time. If such radical-sounding kinds of
parameterization give us the best and most principled overall explanation of the facts,
then we should not be afraid to do it … at least until a better proposal comes to light.
7. Conclusions
I have argued that there are (at least) two ways of assigning morphological case in natural
languages: case can be assigned by a functional head to a nearby NP under agreement
(Chomskian case assignment), and case can be assigned to one NP if there is another NP
in the same local domain (Marantzian case assignment). Languages in which the Case
Dependence of Agreement Parameter of Baker 2008 is set “yes” have (at least some)
Chomskian case assignment; in contrast, languages in which the CDAP is set “no” may
have only Marantzian case assignment. Languages can also have a mixture of the two
modes of case assignment, the Sakha language being a case in point. I then went on to
conjecture that structural ergative case is always a Marantzian dependent case, hence it is
only found in CDAP=no languages or languages without agreement, and I presented
preliminary typological evidence that this might be true. This approach makes the
seemingly radical prediction that, since ergative case and accusative case are sensitive to
the same local domains (phases), the definiteness/position of the object can affect both,
and the tense-aspect of the clause can affect both. However, there is supporting evidence
for this within my initial set of case studies. Finally, and perhaps most radically, I
suggested that languages differ as to whether NPs in the lower case domain (VP) are
automatically included in the higher case domain (CP) or not. The latter sort of language
will be prone to differential object marking and split ergativity; the former will have
uniform case assignment across virtually all clause types. This then is a survey of some
types of variation and parameterization that we seem to find in systems of morphological
case, within current generative terms.
References
Amberber, Mengistu. 2005. Differential subject marking in Amharic. In Competition and
variation in natural languages: the case for case, eds. Mengistu Amberber and
Helen de Hoop, 295-319. Amsterdam: Elsvier.
Anderson, Stephen. 2005. Aspects of the theory of clitics. New York: Oxford University
Press.
22
Baker, Mark. 2008. The syntax of agreement and concord. Cambridge: Cambridge
University Press.
Baker, Mark, and Vinokurova, Nadezhda. 2009. On agentive nominalizations and how
they differ from event nominalizations. Language 85:517-556.
Baker, Mark. 2010. Formal generative typology. In Oxford Handbook of Linguistic
Analysis, eds. Bernd Heine and Heiko Narrog. New York: Oxford University
Press.
Baker, Mark, and Vinokurova, Nadezhda. 2010. Two modalities of case assignment in
Sakha. Natural Language and Linguistic Theory 28:593-642.
Blake, Barry. 1994. Case. Cambridge: Cambridge University Press.
Bok-Bennema, Reineke. 1991. Case and agreement in Inuit. Berlin: Foris.
Bresnan, Joan, and Kanerva, Joni. 1989. Locative inversion in Chichewa: a case study of
factorization in grammar. Linguistic Inquiry 20:1-50.
Chomsky, Noam. 2000. Minimalist inquiries: the framework. In Step by Step, eds. Roger
Martin, David Michaels and Juan Uriagereka, 89-155. Cambridge, Mass.: MIT
Press.
Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: A life in language, ed.
Michael Kenstowicz, 1-52. Cambridge, Mass.: MIT Press.
Cole, Peter. 1985. Imbabura Quechua. London: Croom Helm.
Comrie, Bernard. 2005. Alignment of case marking of full noun phrases. In The world
atlas of language structures, eds. Martin Haspelmath, Matthew Dryer, David Gil
and Bernard Comrie, 398-403. Oxford: Oxford University Press.
Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT Press.
Dixon, R. M. W. 1994. Ergativity. Cambridge: Cambridge University Press.
Dryer, Matthew. 2005. Relationship between the order of object and verb and the order of
adposition and noun phrase. In The world atlas of language structures, eds.
Martin Haspelmath, Matthew Dryer, David Gil and Bernard Comrie, 386-389.
New York: Oxford University Press.
Dunn, John Asher. 1995. Sm'algyax: a reference dictionary and grammar for Coast
Tsimshian. Seattle: University of Washington Press.
Fox, Danny, and Pesetsky, David. 2004. Cyclic linearization of syntactic structure.
Theoretical Linguistics 31:1-46.
Frank, Paul. 1990. Ika syntax. Arlington, Texas: SIL and University of Texas-Arlington.
Gulya, János. 1966. Eastern Ostyak cherstomathy. Bloomington, Ind.: University of
Indiana Press.
Haspelmath, Martin. 1993. A grammar of Lezgian. Berlin: Mouton de Gruyter.
Hopper, Paul, and Thompson, Sandra. 1980. Transitivity in grammar and discourse.
Language 56:251-299.
Hutchison, John. 1981. A reference grammar of the Kanuri language. Madison, Wisc:
African studies program, University of Wisconsin.
Kayne, Richard. 1994. The antisymmetry of syntax. Cambridge, Mass.: MIT Press.
Lefebvre, Claire, and Muysken, Pieter. 1988. Mixed categories: nominalizations in
Quechua. Dordrecht: Kluwer.
Lorimer, D.L.R. 1935. The Burushaski language 1: Introduction and Grammar.vol. 1.
Cambridge, Mass.: Harvard University Press.
23
Marantz, Alec. 1991. Case and licensing. Paper presented at The 8th Eastern States
Conference on Linguistics, University of Maryland, Baltimore.
Mohanan, Tara. 1995. Woodhood and lexicality: noun incorporation in Hindi. Natural
Language and Linguistic Theory 13:75-134.
Moore, John. 1998. Turkish copy-raising and A-chain locality. Natural Language and
Linguistic Theory 16:149-189.
Mulder, Jean Gail. 1994. Ergativity in Coast Tsimshian. Berkeley: University of
California Press.
Rude, Noel. 1986. Topicality, transitivity, and the direct object in Nez Perce.
International Journal of American Linguistics 52:124-153.
Rude, Noel. 1988. Ergative, passive, and antipassive in Nez Perce. In Passive and Voice,
ed. M. Shibatani, 547-560. Amsterdam: John Benjamins.
Vinokurova, Nadezhda. 2005. Lexical categories and argument structure: a study with
reference to Sakha, University of Utrecht: Ph.D dissertation.
Willson, Stephen. 1996. Verb agreement and case marking in Burushaski. Work Papers
of the Summer Institute of Linguistics North Dakota 40:1-71.
Woolford, Ellen. 2006. Lexical case, inherent case, and argument structure. Linguistic
Inquiry 37:111-130.
24